The accurate detection of an object in an image may be useful in any number of contexts, from engineering and the visual arts to inventory control and law enforcement. Often, a simple background subtraction is conducted based on two or more images in which an object of interest has moved relative to the background or is alternately present and absent. Typically, background subtraction is made on a pixel-by-pixel basis. However, where an object and the background are similar in color and/or intensity, simple background subtraction often inaccurately expands the object's boundaries to include portions of the background. Often, such similarities in color and/or intensity are attributable to the presence of the object itself, which may cast a shadow or reflect light onto a background. In reflecting light, an object may even impart its own color onto a background.
To combat such effects, simple background subtraction is often conducted with a very high pixel change threshold. This leads to equally unsatisfactory results, however, with portions of the object that are similar in color and/or intensity being inaccurately attributed to the background.
Accordingly, there exists a need in the art to overcome the deficiencies and limitations described hereinabove.